Graspable Pose Detection Approach Based on Multi-view Point Cloud Fusion

In order to solve the problem that the information collected by the monocular camera is incomplete and the graspable poses are not comprehensive, this paper proposes a graspable poses detection approach based on multi-view point cloud fusion. Firstly, the calibration method is proposed to calibrate multiple depth cameras, and the data obtained from different cameras are transformed to that in robot coordinate system for obtaining relatively complete point cloud data. Secondly, a graspable poses detection approach is proposed to generate reliable graspable poses without relying on the object model, and the neural network is trained end-to-end through a large-scale graspable dataset. At the same time, in order to improve the success rate of grasping, reachability is brought into grasping planning. Experimental results finally show that the proposed method is feasible and effective in dealing with grasping problem.